def main():
parser = argparse.ArgumentParser(description="Extract mutations in VCF and save as simple tabulated file")
parser.add_argument("vcf_paths", metavar="PATH", nargs="+", help="The VCF files")
parser.add_argument("-o", dest="out_path", metavar="PATH", help="Output file. Use - for standard output.")
bglogging.add_logging_arguments(self._parser)
args = parser.parse_args()
bglogging.initialize(self.args)
log = bglogging.get_logger("vcf-to-snvs")
if args.out_path is None:
names = []
for path in args.vcf_paths:
if path != "-":
base_path, name, ext = tsv.split_path(path)
names += [name]
prefix = os.path.commonprefix(*names) if len(names) > 0 else ""
prefix = prefix.rstrip(".")
if len(prefix) == 0:
prefix = "genome"
args.out_path = "{}.tsv.gz".format(prefix)
with tsv.open(args.out_path, "w") as outf:
tsv.write_line(outf, "CHR", "POS", "REF", "ALT")
for path in args.vcf_paths:
log.info("Reading {} ...".format(path))
with tsv.open(path) as inf:
types = (str, str, str, str)
columns = [0, 1, 3, 4]
for fields in tsv.lines(inf, types, columns=columns):
chrom, pos, ref, alt = fields
# ref = ref.upper().strip("N")
# alt = alt.upper().strip("N")
ref_len = len(ref)
alt_len = len(alt)
if ref_len != alt_len or ref_len == 0 or alt_len == 0:
continue
try:
pos = int(pos)
except:
continue
if ref_len == 1:
tsv.write_line(outf, chrom, pos, ref, alt)
else:
for i in range(ref_len):
tsv.write_line(outf, chrom, pos + i, ref[i], alt[i])
def parse_args(self, logger_name): bglogging.add_logging_arguments(self._parser) self.args = self._parser.parse_args() bglogging.initialize(self.args) self.logger = bglogging.get_logger(logger_name) return self.args, self.logger
def main():
parser = argparse.ArgumentParser(
description="Plot training sets statistics")
parser.add_argument("path", metavar="PATH",
help="The statistics json file")
parser.add_argument("-o", dest="out_path", metavar="PATH",
help="The path to save the plot image.")
parser.add_argument("-W", "--width", dest="fig_width", metavar="WIDTH", type=int,
help="The image width.")
parser.add_argument("-H", "--height", dest="fig_height", metavar="HEIGHT", type=int,
help="The image height.")
parser.add_argument("--dpi", dest="fig_dpi", metavar="DPI", type=int, default=100,
help="The image dpi.")
parser.add_argument("-p", "--predictors", dest="predictor_names", metavar="NAMES",
help="The names of the predictors to represent (seppareted by commas).")
parser.add_argument("-i", "--interactive", dest="interactive", action="store_true", default=False,
help="Show the plot in interactive mode.")
bglogging.add_logging_arguments(parser)
args = parser.parse_args()
bglogging.initialize(args)
log = bglogging.get_logger("plot-stats")
log.info("Loading state from {} ...".format(os.path.basename(args.path)))
state = load_weights(args.path)
predictor_names, stats = [state[k] for k in ["predictor_names", "stats"]]
if args.predictor_names is not None:
valid_names = set(predictor_names)
args.predictor_names = [s.strip() for s in args.predictor_names.split(",")]
predictor_names = [name for name in args.predictor_names if name in valid_names]
if len(predictor_names) == 0:
log.error("No scores selected. Please choose between: {}".format(", ".join(valid_names)))
exit(-1)
#log.info("Plotting ...")
fig = plt.figure(figsize=(args.fig_width or 12, args.fig_height or 10.4), dpi=args.fig_dpi or 100)
alpha = 0.7
num_predictors = len(predictor_names)
for i in range(num_predictors):
predictor_name = predictor_names[i]
predictor_stats = stats[predictor_name]
(intervals, dp, dn, cump, cumn, cdp, cdn, tp, tn, fp, fn, mcc, accuracy, cutoff) = [
predictor_stats[k] for k in [
"values", "dp", "dn", "cump", "cumn", "cdp", "cdn", "tp", "tn", "fp", "fn", "mcc", "acc", "cutoff"]]
dax = fig.add_subplot(4, num_predictors, i + 1, title="{}".format(predictor_name))
cdax = fig.add_subplot(4, num_predictors, 1 * num_predictors + i + 1)
tfax = fig.add_subplot(4, num_predictors, 2 * num_predictors + i + 1)
aax = fig.add_subplot(4, num_predictors, 3 * num_predictors + i + 1)
# distribution
dax.grid()
dax.plot(intervals, arrdiv(dp, max(dp)), "r-", alpha=alpha)
dax.plot(intervals, arrdiv(dn, max(dn)), "b-", alpha=alpha)
dax.plot([cutoff, cutoff], [0.0, 1.0], "k--")
dax.legend(('POS', 'NEG'), 'upper center', ncol=2, frameon=False, prop={'size':10})
# cummulative distribution
cdax.grid()
cdax.plot(intervals, arrdiv(cdp, cump), "r-", alpha=alpha)
cdax.plot(intervals, arrdiv(cdn, cumn), "b-", alpha=alpha)
cdax.plot([cutoff, cutoff], [0.0, 1.0], "k--")
cdax.legend(('POS', 'NEG'), 'upper center', ncol=2, frameon=False, prop={'size':10})
# TP/FN/FP/TN
tfax.grid()
tfax.plot(intervals, arrdiv(tp, cump), "r-", alpha=alpha)
tfax.plot(intervals, arrdiv(fn, cump), "c--", alpha=alpha)
tfax.plot(intervals, arrdiv(fp, cumn), "b-", alpha=alpha)
tfax.plot(intervals, arrdiv(tn, cumn), "m--", alpha=alpha)
tfax.plot([cutoff, cutoff], [0.0, 1.0], "k--")
tfax.legend(('TP', 'FN', 'FP', 'TN'), 'upper center', ncol=4, frameon=False, prop={'size':8})
# MCC/Accuracy
aax.grid()
aax.plot(intervals, mcc, "g-", alpha=alpha)
aax.plot(intervals, accuracy, "y-", alpha=alpha)
aax.plot([cutoff, cutoff], [0.0, 1.0], "k--")
aax.legend(('MCC', 'Accuracy'), 'upper center', ncol=2, frameon=False, prop={'size':10})
if args.out_path is not None:
from matplotlib import pylab
log.info("Saving image into {} ...".format(os.path.basename(args.out_path)))
pylab.savefig(args.out_path, bbox_inches=0)
if args.interactive:
plt.show()
def main():
parser = argparse.ArgumentParser(
description="Calculate weights")
parser.add_argument("ranges_path", metavar="RANGES_PATH",
help="JSON file generated with pred-list containing predictors stats. Only min and max are used.")
parser.add_argument("training_path", metavar="TRAINING_PATH",
help="The training set scores. ID column should be POS/NEG for positive/negative sets.")
parser.add_argument("-o", dest="out_path", metavar="OUT_PATH",
help="The file where weights will be saved. Use - for standard output.")
parser.add_argument("-p", "--predictors", dest="predictors", metavar="PREDICTORS",
help="Comma separated list of predictors to fetch")
parser.add_argument("-P", "--precision", dest="precision", metavar="PRECISSION", type=int, default=3,
help="Distribution precision")
parser.add_argument("-f", "--full-state", dest="full_state", action="store_true", default=False,
help="Save intermediate calculations to allow further exploration and plotting")
bglogging.add_logging_arguments(parser)
args = parser.parse_args()
bglogging.initialize(args)
logger = bglogging.get_logger("weights")
if args.out_path is None:
prefix = os.path.splitext(os.path.basename(args.training_path))[0]
if prefix.endswith("-scores"):
prefix = prefix[:-7]
args.out_path = os.path.join(os.getcwd(), "{}-weights.json".format(prefix))
if args.predictors is not None:
args.predictors = [p.strip() for p in args.predictors.split(",")]
logger.info("Loading ranges from {} ...".format(os.path.basename(args.ranges_path)))
with open(args.ranges_path) as f:
pred_stats = json.load(f)
predictor_range = {}
for pid, pstats in pred_stats.items():
predictor_range[pid] = (pstats["min"], pstats["max"])
logger.info("Reading training set {} ...".format(args.training_path if args.training_path != "-" else "from standard input"))
with tsv.open(args.training_path) as f:
# Select predictors from the available predictors in the dataset or user selection
column_names, column_indices = tsv.header(f)
available_predictors = [c for c in column_names if c not in set(COORD_COLUMNS)]
if args.predictors is None:
predictors = available_predictors
else:
missing_predictors = [p for p in args.predictors if p not in set(available_predictors)]
if len(missing_predictors) > 0:
logger.error("Missing predictors: {}".format(", ".join(missing_predictors)))
exit(-1)
predictors = args.predictors
data = pd.read_csv(args.training_path, sep="\t", index_col=False,
usecols=["ID"] + predictors,
true_values=["POS"], false_values=["NEG"])
data.rename(columns={"ID" : "EVT"}, inplace=True)
# Initialize statistics
logger.info("Initializing metrics ...")
step = 1.0 / 10**args.precision
stats = dict()
state = dict(
predictor_names = predictors,
precision = args.precision,
step = step,
stats = stats)
for predictor in predictors:
d = data[["EVT", predictor]]
d = d[np.isfinite(d.iloc[:, 1])]
nump = d.iloc[:, 0].sum()
numn = d.shape[0] - nump
rmin, rmax = d.iloc[:, 1].min(), d.iloc[:, 1].max()
dim = rmax - rmin
size = int(dim / step) + 1
values = [(x * step) + rmin for x in xrange(size)]
logger.info(" {:10}: p={}, n={}, min={}, max={}, bins={}".format(predictor, nump, numn, rmin, rmax, size))
stats[predictor] = dict(
rmin = rmin,
rmax = rmax,
dim = dim,
values = values,
size = size,
vmin = rmin,
vmax = rmax,
dp = [0] * size,
dn = [0] * size,
cdp = [0] * size,
cdn = [0] * size,
cump = 0,
cumn = 0,
tp = [0] * size,
tn = [0] * size,
fp = [0] * size,
fn = [0] * size,
mcc = [0] * size,
acc = [0] * size,
auc = [0] * size,
cutoff = None,
cutoff_index = None,
cutoff_mcc = None,
cutoff_acc = None,
cutoff_auc = None)
positive_count = data.iloc[:, 0].sum()
negative_count = data.shape[0] - positive_count
logger.info(" TOTAL : positive={}, negative={}".format(positive_count, negative_count))
logger.info("Calculating scores distribution and confusion matrices ...")
logger.info("Calculating cumulative distribution ...")
for predictor in predictors:
predictor_stats = stats[predictor]
dp, dn, cdp, cdn = [predictor_stats[k] for k in ["dp", "dn", "cdp", "cdn"]]
cump = 0
cumn = 0
i = len(dp) - 1
while i >= 0:
cdp[i] = dp[i] + cump
cump += dp[i]
cdn[i] = dn[i] + cumn
cumn += dn[i]
i -= 1
predictor_stats["cump"] = cump
predictor_stats["cumn"] = cumn
logger.info(" {}: cump={}, cumn={}".format(predictor, cump, cumn))
logger.info("Calculating accuracy and cutoff ...")
for predictor in predictors:
predictor_stats = stats[predictor]
values, size, tp, tn, fp, fn, mcc, acc = [predictor_stats[k] for k in [
"values", "size", "tp", "tn", "fp", "fn", "mcc", "acc"]]
cutoff = -1
cutoff_index = -1
best_mcc = -1e6
for i in xrange(size):
try:
#http://en.wikipedia.org/wiki/Matthews_correlation_coefficient
mcc[i] = (tp[i] * tn[i] - fp[i] * fn[i]) / sqrt((tp[i] + fp[i]) * (tp[i] + fn[i]) * (tn[i] + fp[i]) * (tn[i] + fn[i]))
#http://en.wikipedia.org/wiki/Accuracy
acc[i] = (tp[i] + tn[i]) / float(tp[i] + fp[i] + fn[i] + tn[i])
except ZeroDivisionError:
mcc[i] = 0
acc[i] = 0
if mcc[i] > best_mcc:
cutoff = values[i]
cutoff_index = i
best_mcc = mcc[i]
best_acc = max(acc)
predictor_stats["cutoff"] = cutoff
predictor_stats["cutoff_index"] = cutoff_index
predictor_stats["cutoff_mcc"] = best_mcc
predictor_stats["cutoff_acc"] = best_acc
logger.info(" {}: cutoff={:.3f}, mcc={:.2f}, accuracy={:.2f}".format(
predictor, cutoff, best_mcc * 100.0, best_acc * 100.0))
if args.full_state:
logger.info("Saving weights with full state ...")
out_path = args.out_path
save_weights(out_path, state)
else:
logger.info("Saving weights ...")
stats = {}
reduced_state = dict(
predictor_names=state["predictor_names"],
precision=state["precision"],
step=state["step"],
stats=stats)
for predictor in state["predictor_names"]:
predictor_stats = state["stats"][predictor]
stats[predictor] = dict(
rmin=predictor_stats["rmin"],
rmax=predictor_stats["rmax"],
dim=predictor_stats["dim"],
values=predictor_stats["values"],
size=predictor_stats["size"],
cdp=predictor_stats["cdp"],
cdn=predictor_stats["cdn"],
cutoff=predictor_stats["cutoff"],
cutoff_index=predictor_stats["cutoff_index"])
save_weights(args.out_path, reduced_state)
return 0
def main():
parser = argparse.ArgumentParser(
description="Calculate Condel label")
parser.add_argument("db_path", metavar="DB_PATH",
help="Functional scores database")
parser.add_argument("weights_path", metavar="WEIGHTS",
help="File containing the scores weights and cutoffs")
parser.add_argument("-p", "--predictors", dest="predictors", metavar="PREDICTORS",
help="Comma separated list of predictors")
parser.add_argument("-u", "--updated-predictors", dest="updated_predictors", metavar="NAMES",
help="Updated predictor names")
bglogging.add_logging_arguments(parser)
args = parser.parse_args()
bglogging.initialize(args)
log = bglogging.get_logger("calculate-label")
log.info("Opening functional scores database ...")
db = FannsSQLiteDb(args.db_path)
db.open()
log.info("Loading state ...")
state = load_weights(args.weights_path)
avail_predictors, precision, step, stats = [state[k] for k in ["predictor_names", "precision", "step", "stats"]]
if args.predictors is not None:
predictors = [p for p in [p.strip() for p in args.predictors.split(",")] if p in avail_predictors]
if len(predictors) == 0:
log.error("Unknown predictors: {}".format(args.predictors))
log.error("Available predictor names are: {}".format(", ".join(avail_predictors)))
exit(-1)
else:
predictors = avail_predictors
if args.updated_predictors is not None:
updated_predictors = [p.strip() for p in args.updated_predictors.split(",")]
if len(predictors) != len(updated_predictors):
log.error("Number of updated predictors does not match with the list of number of predictors")
exit(-1)
else:
updated_predictors = ["{}_CLASS".format(p.upper()) for p in predictors]
log.info("Available predictors: {}".format(", ".join(avail_predictors)))
log.info("Selected predictors: {}".format(", ".join(predictors)))
for predictor, updated_predictor in zip(predictors, updated_predictors):
log.info("Creating predictor {} ...".format(updated_predictor))
db.add_predictor(updated_predictor, FannsDb.CALCULATED_PREDICTOR_TYPE, source=[predictor])
cutoffs = []
for predictor in predictors:
cutoff, mcc, acc = [stats[predictor][v] for v in ["cutoff", "cutoff_mcc", "cutoff_acc"]]
log.info("{}: cutoff={}, MCC={}, accuracy={}".format(predictor, cutoff, mcc, acc))
cutoffs += [cutoff]
log.info("Calculating ...")
start_time = partial_start_time = time.time()
try:
for num_rows, row in enumerate(db.query_scores(predictors=predictors), start=1):
scores = row["scores"]
d = {}
for i, predictor in enumerate(predictors):
score = scores[predictor]
if score is None:
continue
cutoff = cutoffs[i]
updated_predictor = updated_predictors[i]
d[updated_predictor] = 0.0 if score < cutoff else 1.0
db.update_scores(row["id"], d)
partial_time = time.time() - partial_start_time
if partial_time > 5.0:
partial_start_time = time.time()
elapsed_time = time.time() - start_time
log.debug(" {} rows, {:.1f} rows/second".format(hsize(num_rows), num_rows / elapsed_time))
db.commit()
except KeyboardInterrupt:
log.warn("Interrupted by Ctrl-C")
db.rollback()
except:
db.rollback()
raise
finally:
db.close()
def main():
parser = argparse.ArgumentParser(
description="Plot training sets statistics")
parser.add_argument("path", metavar="PATH",
help="The statistics json file")
parser.add_argument("-o", dest="out_path", metavar="PATH",
help="The path to save the plot image.")
parser.add_argument("-p", "--predictors", dest="predictor_names", metavar="NAMES",
help="The names of the predictors to represent (seppareted by commas).")
parser.add_argument("-i", "--interactive", dest="interactive", action="store_true", default=False,
help="Show the plot in interactive mode.")
bglogging.add_logging_arguments(parser)
args = parser.parse_args()
bglogging.initialize(args)
log = bglogging.get_logger("plot-stats")
log.info("Loading state ...")
state = load_weights(args.path)
predictor_names, stats = [state[k] for k in ["predictor_names", "stats"]]
if args.predictor_names is not None:
valid_names = set(predictor_names)
args.predictor_names = [s.strip() for s in args.predictor_names.split(",")]
predictor_names = [name for name in args.predictor_names if name in valid_names]
if len(predictor_names) == 0:
log.error("No scores selected. Please choose between: {}".format(", ".join(valid_names)))
exit(-1)
log.info("Plotting ...")
fig = plt.figure()
ax = fig.add_subplot(111)
ax.grid()
ax.set_xlabel("False Positive Rate (1 - especificity)")
ax.set_ylabel("True Positive Rate (sensitivity)")
num_predictors = len(predictor_names)
for predictor_name in predictor_names:
predictor_stats = stats[predictor_name]
(size, tp, tn, fp, fn) = [predictor_stats[k] for k in ["size", "tp", "tn", "fp", "fn"]]
tpr = [1.0] * (size + 1)
fpr = [1.0] * (size + 1)
for i in range(size):
tpr[i + 1] = (float(tp[i]) / (tp[i] + fn[i]))
fpr[i + 1] = (float(fp[i]) / (fp[i] + tn[i]))
ax.plot(fpr, tpr, "-")
ax.legend(tuple(predictor_names), "lower right", shadow=False)
ax.plot([0.0, 1.0], [0.0, 1.0], "--", color="0.75")
if args.out_path is not None:
from matplotlib import pylab
pylab.savefig(args.out_path, bbox_inches=0)
if args.interactive:
plt.show()
def main():
parser = argparse.ArgumentParser(
description="Calculate Condel score")
parser.add_argument("db_path", metavar="DB_PATH",
help="Functional scores database")
parser.add_argument("weights_path", metavar="WEIGHTS",
help="File containing the scores weights and cutoffs")
parser.add_argument("-p", "--predictors", dest="predictors", metavar="PREDICTORS",
help="Comma separated list of predictors")
parser.add_argument("-u", "--updated-predictor", dest="updated_predictor", metavar="NAME",
help="Updated predictor name")
bglogging.add_logging_arguments(parser)
args = parser.parse_args()
bglogging.initialize(args)
log = bglogging.get_logger("calculate")
log.info("Opening functional scores database ...")
db = FannsSQLiteDb(args.db_path)
db.open()
updated_predictor = args.updated_predictor or "CONDEL"
predictors = set([p["id"] for p in db.predictors()])
if updated_predictor not in predictors:
log.info(" Creating predictor {} ...".format(updated_predictor))
db.add_predictor(updated_predictor, FannsDb.CALCULATED_PREDICTOR_TYPE, source=predictors)
log.info("Loading state ...")
state = load_weights(args.weights_path)
avail_predictors, precision, step, stats = [state[k] for k in ["predictor_names", "precision", "step", "stats"]]
if args.predictors is not None:
predictors = [p for p in [p.strip() for p in args.predictors.split(",")] if p in avail_predictors]
if len(predictors) == 0:
log.error("Unknown predictors: {}".format(args.predictors))
log.error("Available predictor names are: {}".format(", ".join(avail_predictors)))
exit(-1)
else:
predictors = avail_predictors
log.info("Available predictors: {}".format(", ".join(avail_predictors)))
log.info("Selected predictors: {}".format(", ".join(predictors)))
log.info("Calculating ...")
start_time = partial_start_time = time.time()
try:
for num_rows, row in enumerate(db.query_scores(predictors=predictors), start=1):
scores = row["scores"]
condel = wsum = 0
for predictor, score in scores.items():
if score is None:
continue
predictor_stats = stats[predictor]
rmin, rmax, dim, size, cdp, cdn, cutoff = [predictor_stats[k] for k in [
"rmin", "rmax", "dim", "size", "cdp", "cdn", "cutoff"]]
if predictor in PREDICTOR_TRANSFORM:
score = PREDICTOR_TRANSFORM[predictor](score)
r = (score - rmin) / dim
index = int(r * size) if score < rmax else size - 1
if score < cutoff:
w = 1 - cdn[index]
else:
w = 1 - cdp[index]
wsum += w
condel += w * score
#log.info("{}={}, w={} -> {}".format(predictor_name, score, w, score * w))
if wsum != 0:
condel /= wsum
d = {updated_predictor : condel}
db.update_scores(row["id"], d)
#log.info(">>> CONDEL={}".format(condel))
else:
log.warn("wsum = 0, condel={}, scores={}".format(condel, repr(scores)))
partial_time = time.time() - partial_start_time
if partial_time > 5.0:
partial_start_time = time.time()
elapsed_time = time.time() - start_time
log.debug(" {} rows, {:.1f} rows/second".format(hsize(num_rows), num_rows / elapsed_time))
log.info("Commit ...")
db.commit()
except KeyboardInterrupt:
log.warn("Interrupted by Ctrl-C")
db.rollback()
except:
db.rollback()
raise
finally:
db.close()
def main():
parser = argparse.ArgumentParser(
description="Prepare SNV's dataset from individual training sets")
parser.add_argument("pos_path", metavar="POS_SET",
help="The positive training set file")
parser.add_argument("neg_path", metavar="NEG_SET",
help="The negative training set file")
parser.add_argument("-m", "--map", dest="map_path", metavar="MAP",
help="Optional mapping file for feature id's. Format: DST SRC")
parser.add_argument("-o", dest="out_path", metavar="PATH",
help="Output file. Use - for standard output.")
bglogging.add_logging_arguments(parser)
args = parser.parse_args()
bglogging.initialize(args)
logger = bglogging.get_logger("training-sets")
if args.out_path is None:
prefix = os.path.commonprefix([
os.path.splitext(os.path.basename(args.pos_path))[0],
os.path.splitext(os.path.basename(args.neg_path))[0]])
prefix = prefix.rstrip(".")
args.out_path = os.path.join(os.getcwd(), "{}-training.tsv".format(prefix))
if args.map_path is not None:
logger.info("Loading map ...")
prot_map = {}
with tsv.open(args.map_path) as f:
for dst_feature, src_feature in tsv.lines(f, (str, str)):
if len(src_feature) > 0:
if src_feature not in prot_map:
prot_map[src_feature] = set([dst_feature])
else:
prot_map[src_feature].add(dst_feature)
else:
prot_map = None
logger.info("Processing ...")
hits = dict(POS=0, NEG=0)
fails = dict(POS=0, NEG=0)
start_time = datetime.now()
with tsv.open(args.out_path, "w") as wf:
for event_type, path in (("POS", args.pos_path), ("NEG", args.neg_path)):
logger.info(" [{}] Reading {} ...".format(event_type, path))
with tsv.open(path) as f:
types = (str, int, str, str)
for protein, pos, aa1, aa2 in tsv.lines(f, types):
protein = protein.strip()
if prot_map is not None:
if protein not in prot_map:
logger.debug("[{}] Unmapped protein: {}".format(event_type, protein))
fails[event_type] += 1
continue
proteins = prot_map[protein]
else:
proteins = [protein]
hits[event_type] += 1
for p in proteins:
tsv.write_line(wf, p, pos, aa1.strip(), aa2.strip(), event_type)
logger.info(" POS NEG")
logger.info("SNVs {POS:>8} {NEG:>8}".format(**hits))
if args.map_path is not None:
logger.info("unmapped {POS:>8} {NEG:>8}".format(**fails))
logger.info("Finished. Elapsed time: {}".format(datetime.now() - start_time))